Abstract: Aiming at the two difficulties in characteristic index digging of combat system of systems (CSoS), namely operation data generation and digging method selection, this paper proposes a new digging method, that is, using the simulation testbed to generate operation data, then adopting the machine learning to dig characteristic index. Two methods of characteristic index digging based on machine learning are researched: (1) the method based on network convergence, divides the communities for fundamental indexes based on their relationship, and obtains the characteristic indexes by principal component analysis (PCA); this method is applied to dig the characteristic indexes of air defense ability. (2) the method based on ensemble learning, generates test data by bagging, trains model by CART decision trees, and obtains the characteristic indexes by PCA; this method is applied to dig the characteristic indexes of air defense breakthrough ability.

Key words: combat system of systems (SoS), capability, characteristic index digging, simulation testbed, machine learning